Department o f Economics

Working Paper 14/89

PRICE AND QUANTITY MEASUREMENTS: THEORETICAL BIASES I N EMPIRICAL PROCEDURES

Jonathan N i tzan

Department o f Economics Working Papers a re d r a f t s o f work i n progress and should no t be c i t e d wi thout the permission of the author. Requests fo r a d d i t i o n a l copies should be addressed to:

Working Papers Coordinator Department o f Economics McGi l l U n i v e r s i t y 855 Sherbrooke St. West Montreal, Quebec H3A 2T7

The f i n a n c i a l ass is tance o f l e fonds FCAR pour l ' a i d e e t l e sou t i en h l a recherche i s g r a t e f u l l y acknowledged.

ABSTRACT

The presence of qualitative changes in the nature of many commodities hinders our ability to construct meaningful price and quantity indices, This paper assesses some of the quality-change literature that seeks to resolve this problem. Several writers have endeavoured to develop objective, theory-neutral procedures designed to measure qualitative changes in some timeless, quantitative units, These measures, they argued, could be used to properly adjust ordinary price and quantity statistics for distortions introduced by quality changes- A careful examination suggests, however, that such procedures are neither objective nor free of theoretical biases. First, all existing attempts to develop 'objective' commodity measures in the presence of quality change are besieged by a constant resort - explicit or implicit- to 'subjective' considerations, Second, both the idea that

quality can be measured and the methods developed for that purpose are closely tied with the neoclassical theoretical paradigm, particularly with its emphasis on perfect competition and equilibrium,

La presence de chanqements qualitatifs dans la nature de nombreux biens nous empsche da&tablir des indices significatifs des prix et des quantites, Laauteur de cet article etudie plusieurs documents sur les chanqements de qualite qui cherchent a resoudre ce probl&me, Plusieurs auteurs ont cherche &laborer des procedures objectives et thkriquement neutres pour mesurer les chanqements qualitatifs dans cetaines unites quantitative5 eternelles, Ces mesures, soutiennent-ils, devraient pouvoir servir a ajuster comme i l se doit les indices des prix et des quantites en fonction des distorsions trees par les chanqements de qualit&, Une analyse minutieuse de ces mesures nous r&v&le qu'elles ne sont ni objectives ni exemptes de partialite thkrique, En premier lieu toute tentative d'etablir des mesures <<objectives>> des biens en presence de chanqements de qualit& est assortie d a u n recours constant -- explicite ou iarplicite -- a des considerations daordre <<subjectif>>, En second lieu, l'idk que la qualite peut se mesurer et les methodes d&velopp&es a cette fin sont etroitement liees au paradigme thkrique n&oclassique, notamment l'importance qu'il attache a la concurrence parfaite et a l'muilibre,

Introduction

Economists continuously argue about prices and quantities of commodities. The debate focuses

mainly on why these attributes change and less on how they change. Of course, there is disagreement on

the latter question, but this mostly pertains to the adequacy of various measurements for different theoretical

problems. When considered in isolation, the measurement of prices and quantities is commonly viewed as

an empirical, objective procedure. The scientific method requires that theories be tested against objective

data and hence the measurement of such data must, whenever possible, be sufficiently independent of the

theoretical debate itself. For instance, an increase in the list price of passenger cars can be explained by

changes in tastes or technology, which affect demand and supply in a perfectly competitive market. It could

also be rationalized by resorting to changes in producer mark-ups under conditions of oligopoly. However,

both theoretical approaches are trying to explain changes in the same statistical price series. Similarly, a

theory emphasizing rational expectations can argue that the business cycle results from a particular time-

series process, while a theory that stresses institutional aspects can explain the cycle by changes in the rate

of profit. However, proponents of both theories will probably use the same statistical series for real GNP as

their principal variable of interest. List prices of passenger cars in the first example and real GNP figures in

the second illustration are deemed adequate precisely because their measurement is believed to be

sufficiently independent of the corresponding theoretical debates.

The presumption of theory-neutral measurement seems adequate in simple abstract cases when the

'commodity' being considered remains unaltered. In such cases, it is then sufficient to count how many units

were produced in each period and to observe the prices at which they were sold. For instance, suppose

Ford Motors produced 100,000 Mustang cars at a unit price of $10,000 in 1975 and manufactured 150,000

units at a price of $14,000 per car in 1985. If we can presume that the Mustang of 1975 was identical to the

one produced in 1985, we can, without ever defining what a Mustang is, conclude that there was a 50 per

cent increase in quantity and a 40 per cent rise in price. On the other hand, if we acknowledge that the two

models are different, such direct comparison has little meaning and we must now both define the

'commodity' and describe how it changes over time. The two Mustang models may vary in aspects of

production - such as the technology with which they were manufactured, the labour invdved in their

assembly, and their material composition. They could also vary in their so-called 'consumption attributes' -

- such as weight, size, power, shape, speed, comfort, cdour, fuel efficiency, noise and chemical pollution.

Under such circumstances, we must somehow denominate all such 'quality' differences in universal,

quantitative terms and adjust our computations accordingly. For instance, if because of such changes, a

1985 modd contained twice as much 'automobile quality' as the 1975 model, we would have a 200 per cent

rise in quantity produced and a 30 per cent decrease - not increase - in unit price! On the other hand, if

quality was found to be 50 per cent lower in the 1985 modd than in the 1975 one, we would end up with

a 180 per cent rise in price and a 25 per cent reduction in quantity!'

Clearly, whenever the nature of the commodity changes, the measurement of such changes in

'quality' is crucial for price and quantity calculations. But there then arises the questions of how to measure

quality and whether such measurements can remain free of theoretical considerations. This problem is

clearly exacerbated as we move from a single commodity to wider aggregates. In devising output and price

indices for the entire car industry, for instance, we must also account for the addition of new models,

deletion of d d ones and temporal changes in the industry's product mix. If we move to even broader

indices such as real investment and its price deflator, or real GNP and the GNP deflator, our difficulties

propagate since the concept of 'aggregate quality' is even more elusive.

The complex issue of comparing different commodities in time and space has occupied economists

since the days of Adam Smith. In fact, the need to convert qualitative aspects into quantitative magnitudes

of 'invariant' nature relates to epistemological problems of cognition, consciousness, subjectivity and

objectivity, which have troubled some of the greatest philosophers since Plato and Aristotle. The

predicament only intensified with the rise of capitalism, and prominent thinkers (like Hume, Kant, Hegel and

Mam) dealt with them extensively. However, such difficulties remained largely theoretical until the present

century, when the volume of quantitative statistical data grew substantially. With the evolution of national

accounts after the 1930s, there was a pressing need for methods of estimation and measurement that would

overcome the problem of historical change. As statistical bureaucracies expanded, techniques were

formalized and the dder theoretical debates gradually gave way to 'objective' procedures, presumably

untainted by pseudo-scientific or philosophical disputes.

The purpose of this paper is to assess some of the literature that sought to resolve problems in

commodity measurement, specifically those posed by quality change. Explicit points of controversy in this

literature have received considerable attention and we do not attempt to provide yet another summary of

these debates. Instead, we direct our examinations toward some fundamental aspects which remained

largely unexplored in the literature. We demonstrate that, despite its claims, the literature has failed to

produce theory-neutral measurement methods. Current procedures are apparently based on specific theories

of price determination, illqualified assumptions about market structure and broader views on the motivating

forces in modem society. These bases of measurement are highly significant but, unfortunately, they are

rarely stated explicitly when the data are produced by statistical agencies and when data are used by

economists.

Commodities and their Characteristics: Search for Obiective Criteria

Because they seem to change, commodities themselves cannot be used as a uniform standard for

temporal comparison and alternative units must be sought. One solution for this difficulty has been to define

commodities as collections of smaller building blocks, or 'characteristics,' which are readily measurable.

Lancaster (1971) for instance, formalized a taxonomy of characteristics as a basis for a new approach to

consumer demand (although this could be equally used in reference to non-consumer goods). He proposed

to view the relationship between people and things as a two-staged affair consisting of an objective

relationship between things (commodities) and their characteristics and a subjective relationship between

characteristics and people:

All goods possess objective characteristics ... The relationship between a given quantity of a good (or a collection of goods) and the characteristics which it possesses is essentially a technical relationship, depending on the objective properties of the goods and, sometimes, a context of technological 'know-how' as to what the goods can do, and how. l ndividuals differ in their reactions to different characteristics, rather than in their assessment of the characteristics content of various good collections. (p. 7, emphases added)

Clearly, in order to identify objective characteristics, we must effectively demarcate them from

subjective interpretations. This separation is of paramount significance but then we may ask on what

practical criteria it should be based? For example, how do we know that an apparent attribute of a medicine

is an objective characteristic and not simply our personal conviction? Lancaster's solution is straightforward:

Operationally speaking, it is universality that is important. If weryone believes that snake oil has special medical properties, we would analyze behaviour as though this were indeed true even if, in some objective sense, it could not be shown to be true; or even if it could be shown to be false, provided the negative proof was unknown or unaccepted in the society being studied. (p. 18-9, emphases added)

This method is seriously flawed in a number of ways. First, it embodies a double standard for objectivity

which may lead to logical inconsistency: if a characteristlc was deemed to be objectively true by a criterion

of unanimous consent, how could such a characteristic be shown to be objectively false at the same time?

Second, the need for 'universal acceptance' is a highly demanding because wen minor dissent will render

it non-operational. Third and most important, if 'acceptance' and 'belief' are criteria for objectivity what

should be the criteria for subjectivity' This latter point deserves further elaboration.

In a modem industrial society, there is abundant information about commodities, some of which

appears to be 'generally accepted.' Yet, one may still ponder on how such general acceptance makes the

information 'objective.' To illustrate some of the attendant difficulty, consider for instance Lancaster's simple

diet example (p. 17) where he tabulates the 'consumption technology' that relates different food items such

as milk, eggs, and meat, with some nutrient characteristics like calories, protein and vitamins. The amount

of each nutrient per unit of food is said to be objective because it was measured by 'official agencies,' but

what happens if such agencies err in their measurements or intentionally falsify the data? This assertion

demonstrates how the 'objective' characteristics of food may depend on the social organization of

information. Let us carry this example one step further and consider the possibility of scientific progress,

where new theories or methods of measurements may produce different numbers of nutrient contents.

Could we permit the objective method for measuring protein in milk, for instance, to alter over time, or

should we adopt a single method of measurement for all periods? Lancaster provides no clues as for how

to address such dynamic historical questions.

While the accuracy of official data and stability of measurement methods in natural sciences may

seem to present little practical difficulty, they nevertheless polnt to the potential hazard in relying on

'universality' as a criterion for 'objectivity.' The difficulties are no longer hypothetical, however, when we

consider information that is made universally accepted not by 'official agencies' but via means of mass

persuasion. In Hwley's Brave New World, genetic engineering and brain washing are used to create legions

of twin consumers that share not only the same preferences but also an identical perception of reality.

Clearly, one does not have to go as far as Hdey's black utopia to identify the co-influence stemming from

various media of persuasion. When advertisement argues that vitamin pills enhance our long term health,

for instance, it concurrently affects our perception of what these vitamins can do and our preference for

them. Similarly, when an advertisement encourages us to consume artificial sweeteners but fails to identify

their possible link to cancer, it may influence both our preferences and knowledge at the same time. The

existence of such ambiguities greatly reduce our ability to use 'universal acceptability' as a criterion for

separating objective characteristics from subjective preferences.

All of these questions belong to an age-long and unresolved philosophical debate on the nature of

subject and object which, surprisingly, is not acknowledged by Lanca~ter.~ He simply asserts that because

characteristics are objective, it is obvious that their measurement is 'simply a technical matter' which

presents no 'operational problems' and should be resolved by engineers (p. 115).' The difficulty lies

elsewhere, namely in determining the relevant characteristics.

The Choice of Relevant Characteristics

Complete characterization of commodities is often impractical according to Lancaster, simply

because the number of characteristics may be too large. However, we can still find an adequate

approximation with a partial description that ignores immaterial aspects of a commodity, provided such

description encompasses all of the relevant facets of quality. The natural question that arises in this context

is how can we distinguish between 'relevant' and 'irrelevant' characteristics? Lancaster stipulates that a

characteristic is relevant to a situation if ignoring its existence would lead to different predictions about

people preferences and choice. But, the bases of predictions are subjective. They must rely on a choice

among theories and so Lancaster's recipe for the concrete description of commodities is no longer theory-

neutral.

An example of the intrusions of such theoretical considerations is provided by Triplett (1 983) as part

of a broad distinction between input and output characteristics:'

... quality variation in an input exists if substitution of different varieties or examples of this input creates variations in output or cost that are not explained by the factors included in the production or cost function. A quality Is an input characteristlc if it reduces that unexplained variation. (p. 279)

Similarly,

Something is an output characteristic if it accounts for, or partly accounts for, the unexplained variation in resource usage occasioned by changes in the varieties of nonhomogenous goods produced. (p. 294)

To illustrate Triplett's approach, consider the case of personal computers. 'Computer speed' should be

regarded as a relevant input characteristic if variations in speed help to explain changes in the output or

'user-value' generated by the computer. On the other hand, computer speed should not be viewed as a

relevant output characteristics if increased speed does not seem to entail higher production cost or an

additional use of resources in producing the computer. Another example arises from the consideration of

automobiles. 'Car size' is not an input characteristic if it does not appear to affect consumer preferences or

utility but it is an output characteristic if it has an apparent impact on production cost.

Two questions arise in this context. First, one may ask which methods -- 'user value' for input,

'resource-cost' for output, or both - should be used to identify relevant characteristics for any particular

commodity7 According to Triplett (p. 305) it does not matter. Under conditions of perfect competition and

equilibrium, the two methods yield the same numbers because 'the marginal cost of producing a quality

change must approximate the incremental value of it to the user.' This must be so, wries Triplett, because

otherwise 'a reallocation of resources would take place.'

These conclusions, of course, are valid only to the extent one accepts the theoretical notions of

efficient allocation and equilibrium together with a presumption of perfect competition. Indeed, Triplett

acknowledges that in 'reality' the two methods would usually yidd different measures for quality because

there are 'shifts in functions, interference with competitive allocation, or wrong data' (ibid.). The prerequisite

of 'competitive equilibrium' in this context has been repeatedly emphasized by most commentators in the

quality change debate. Unfortunately, these commentators have generally failed to provide guidance for

identifying characteristics when commodities are produced and traded in alternative market structures or

in disequi~ibrium.~

A second problem arises because Triplett's recipe is simply incomplete and cannot be readily

applied. The procedure requires that we identify the way in which characteristics affect output or input.

Unfortunately, such causal relationships cannot simply be 'observed' and must be based, at least in part,

on economic theory. Furthermore, given a variety of competing theoretical perspectives, the identification

of relevant characteristics hinges on the particular choice of theory. In our earlier computer illustration, for

instance, we need to chose a specific production theory of computer services (to identify input

characteristics) and a cost theory of computer manufacturing (to identify output characteristics).

Unfortunately, Triplett's procedure provide no guidance on how to select the 'appropriate' theory that should

be used in each case.

This latter point is significant for measurement because if distinct theories generate different sets of

relevant characteristics they also lead to different measurements of quality. For instance, consider the

relation between the required duration of a bachelor programme in economics and the overall 'quality' of

the programme in producing graduates. If years of schooling are believed to enhances the 'human capital'

of programme participants (in other words, if 'duration' is taken as a relevant input characteristic), an

increase from three to four years in the duration of the programme should be considered as quality

improvement. On the other hand, if programme duration is regarded as an irrelevant input characteristics,

such an increase obviously does not affect the programme's quality. Yet a third perspective may hold that

programme duration is a relevant input characteristic because it undermines the production of human

capital in graduates. From this perspective, increase in duration is definitely a quality deterioration! Evidently,

each theoretical perspective in this context implies a different measure for quality change. The choice of one

particular measure must then depend on our theoretical preferences and hence cannot be considered

'obje~tive.'~

Triplett, like other participants in the quality change literature, does not acknowledge the potential

impact of such theoretical diversity on quality measurement. As noted above, this literature generally accepts

the neoclassical foundations of perfect competition and equilibrium, while alternative views are simply not

considered. Furthermore, even within this limited framework, it is customary to talk about the production, I

or the util0ky function which must be empirically 'discovered' as if competing formulations cannot coexist.'

Unfortunately, even under this imposed theoretical uniformity, the precise functions for consumption,

production, cost and utility are admittedly unknown and practical methods of measurement always rely on

ad hoc procedures.

S~ecification Pricina and Qualitv Chanae

In practice, most price indices (and associated quantity measures) are based on what is known as

'specification pricing,' where each commodity in the index must fit into some predetermined range of

product attributes or 'specifications.' In this way, comparisons are restricted to only those commodities

whose specifications remain unchanged over the period and a 'pure' price change can be thus assessed.

An important shortcoming of this method stems from its inability to properly account for quality change.

Indeed, the Price Statistics Review Committee (headed by Stigler in 1961) was concerned that adherence

to strict specifications, in the presence of continuous changes in product mix and commodity quality, would

greatly reduce the relevance of the index. As a partial remedy, the Committee recommended we adopt

'more general rules for recognition of comparable qualities' and restructure specifications along 'more

flexible lines' (p. 34). Following this suggestion, U.S. agencies, such as the Bureau of Labor Statistics,

developed 'subspecifications' within main specifications. These extensions provided greater 'flexibility' in

fitting products into existing descriptions but they still did not resolve the pending problem of quality change.

Two points of difficulty deserve attention here.

The United Nations' Guidelines

First, the rules governing the development of specifications are obscure. The United Nations'

Guidelines on Principles of a System of Price and Quantity Statistics do not explain how to distinguish

between objective variations in quality and differences in subjective interpretations. In fact, the Guidelines

allow national statisticians considerable freedom in their choice of characteristics:

In the case of goods, all differences in physical composition, components, size style, packaging and operating characteristics (for example, capacity, power, speed, durability etc.) should be considered quality differences ... In the case of services, qualtty characteristics relate to such attributes as the activities constituting the services, the conditions under which the services are rendered, the level of skill and training of the persons rendering the services and, if feasible, the benefit generally expected from the services. (p. 9, section 45, emphasis added)

This latitude in choosing relevant characteristics is somewhat restricted along lines discussed in our previous

section:

As an exception to the rule stated above, differences in the circumstances of production that do not lead to differences in utilization or function should not generally be considered quality differences. (p. 9, section 46)

Unfortunately, the United Nations does not provide guidance for the practical implementation of this latter

exception. In the absence of clear criteria, the choice of 'relevant' characteristics (for input or output) to be

included in specifications is left to the discretion of statistical-service officials and naturally, arbitrary

definitions of quality lead to arbitrary conclusions about changes in quality. The ambiguity is further

enhanced because most statistical services do not provide adequate narratives for their commodity

specifications.'

A second difficulty arises in bridging the operational gap between fixed specifications and temporal

changes in quality. Even with many layers of 'objective' subspecification, the overall description of

commodities in a price index is ultimately given and hence does not allow for variations arising from quality

change in existing commodities, introduction of new commodities and disappearance of old ones. Most

statistical services attempt to overcome these problems and preserve the continuity and validity of

specification indices, but procedures employed for this purpose, such as ones described in the United

Nations' Guidelines, contain strong theoretical biases, particularly toward conventional neoclassical price

theory and the presumption of equilibrium.

Under specification pricing, a quality change is said to occur when the product or service no longer

fits into the predetermined specification. The extent of the change can vary from a minor modification

affecting a single aspect in the specification (such as a 2 per cent increase in 'car speed') to a multifaceted

metamorphosis (such as a replacement of an old automobile model by a new one). The problem, of course,

is how to quantify such changes in some universal units of 'quality.' For this, we need to know the relative

contribution of each characteristic to the overall quality of the commodity; that is, we must know the

'consumption technology' or the 'production function,' whatever the case may be. Unfortunately, these

structures are unknown so methods of indirect estimation are substituted for direct measurements of quality

change.

The common adjustment procedure of statisticians uses observed changes in cost or price to

estimate non-observable changes in quality. The U.S. Bureau of Labor Statistics, for instance, relies on cost

informatlon furnished by producers to estimate quality changes in the Producer Price Index. For instance,

if Westinghouse increases the BTU output of an air-conditioning model and informs the Bureau that this

modification requires a 15 per cent rise in cost (under the original technology and factor prices), the change

is taken to represent a 15 per cent improvement in quality.' The same principle is applied in 'splicing,' a

method used when a new commodity replaces an old one. As a hypothetical illustration for splicing,

consider the introduction of Sony compact disc (CD) players priced at $400, as potential replacement for

turntables of an earlier technology which cost $200 per unit. When it is decided to replace the turntable by

the CD player in the Consumer Price Index, the observed price difference is interpreted as a 100 per cent

increase in quality (namely, in quantity of 'music machines' produced), while the CPI itself remains

unaffected.'' The rationale behind these methods is, again, very simple. It is presumed that in equilibrium,

under conditions of perfect competition, the ratio of qualities between two varieties in time or place is just

equal to the corresponding ratio of prices." But then this explanation raises two difficult questions.

Diseauilibrium and Market Im~erfectiong

The first difficulty stems from the possibility of disequilibrium. When markets are out of equilibrium,

there is no unique mapping between qualities and prices and the popular method simply breaks down. To

overcome this obstacle in the case of splicing for instance, the United Nations' Guidelines (p. 10)

recommend we replace products in a price index when the assumption that price differences between the

two products are proportional to quality differences is 'most likely to be true.' In simple words, statisticians

are advised to perform splicing when markets are in equilibrium. Unfortunately, criteria for identifying

occurrences of equilibrium are yet to be developed and the Guidelines concede there is a 'difficulty' here.

So how should the quality of air conditioners, 'music machines' or any other commodity be adjusted in the

meantime? The United Nations have no solution and admit that such corrections must be 'essentially

pragmatic' (p. 10).

The second complication emerges when commodities are produced and exchanged in 'imperfect'

markets. Incidence of government intervention, oligopdistic practices, or non arms-length transactions could

distort or completely destroy the functional relations between price and quality which are presumed to exist

under 'perfect' competition. Again the method breaks down. Surprisingly, the United Nations' Guidelines pay

almost no attention to these potential hazards. But although only two cases of market 'imperfections' are

discussed, these are sufficient to illustrate the ad hoc nature of proposed solutions.

One case involves the mandatory installation of anti-pollution equipment on automobiles. Here the

Guidelines recommend to treat such changes as quality improvements and evaluate them by their cost of

production. The explanation for this recommendation, however, is highly confused. The United Nations agree

that because consumers and producers are given no choice in this matter, changes in cost may not provide

an accurate measure for quality improvement. Nevertheless,

... it is still appropriate to treat the required improvements as increases in quality, not price. Price and quantity statistics are intended to provide objective measures (in so far as these are possible) of what has happened to output and its price. Measuring the utility or welfare that output yield is beyond their scope. (p. 10, emphasis added)

But then,

It is, of course, true that the very concept of "output" involves some assumption about utility; however, certain conventions have been adopted for use in national accounting and it seems preferable to maintain the same conventions here. (ibid., emphasis added)

So despite elaborate theoretical attempts to resolve the issue 'once and for all,' it appears that the

philosophical questions of objectivity and subjectivity persist unabated. Within the framework of equilibrium

and perfect competition, these issues are conveniently ignored but outside this framework one must resort

to 'conventions.'

The Guidelines also identify another 'special case of d'Mculty' invdving internal transactions

between related enterprises or branches of the same company. Since prices set under these conditions may

be 'quite arbitrary,' the United Nations (p. 12) suggests we 'abandon value as one of the primary measures'

and replace it with 'a measure of physical quantity,' combined with an estimate of 'what the equivalent

market price would have been.' This recommendation is puzzling for two reasons. First, how could we

obtain direct measures of physical quantity when, in the presence of quality changes, such measures were

deemed infeasible to begin with? Second, what formula should be used to impute hypothetical market prices

and what is the validity of such imaginary prices? The Guidelines do not explore these questions.

This commentary is significant because the practical estimation of quality change appears to rely

exhaustively on the assumption of equilibrium and perfect competition. In reality, however, almost every

market suffers from some sort of 'imperfection' and experiences of rapid dynamic change seem contrary

to the notions of stability and equilibrium. Even simple examples can illustrate the remarkable disparity

between neoclassical assumptions and underlying realities in this context. Consider, for instance, quality

changes in the nature of military hardware purchased by governments. In order to adjust the price deflator

for military spending to such changes, the U.S. Bureau of Economic Analysis adopts the usual methodology

of 'production cost' and 'splicing.' However, weapon systems are not produced and sold in a perfectly

competitive market but, rather, in an environment of monopsony and oligopoly from the buyer and seller

sides, respectively. Consequently, the meaning of 'equilibrium' in such a market is unclear. It is also not clear

what 'utility function' the government attempts to maximize. The 'cost function' introduces further ambiguities

because, in the reality of military procurement, it is often hard to establish the meaning of 'normal profits'

and even to separate genuine cost from concealed eamings.12

Questions raised by thesg difficulties are by no means limited to military commodities. Can markets

for automobiles, consumer electronics, civilian aircraft, medical equipment, machine tools and ships, for

instance, be considered perfectly competitive? What is the level of 'normal profit' to be included in cost

functions for industries producing for these markets? What are the 'utility functions' for consumers in these

markets? How should we interpret attempts to influence consumer preferences in each of these cases? What

is the impact of government intervention (through regulation, taxes and subsidies for instance) on prices in

such markets? These questions must be clearly answered before we can interpret the validity of quality

adjustments to commodities produced in each market. To our knowledge, however, these issues have not

been sufficiently addressed in the quality-change literature. Rather than deal with such basic questions,

many researchers have chosen to cultivate and refine existing approaches. We deal with one of these

attempts in the following section.

Hedonic Rearessions for Price Indices

The most systematic method of adjustment for quality change can be found in the application of so-

called 'hedonic regressions.' The method was first suggested and applied by Court (1939) and was latter

used in studies by Stone (1 956), Griliches (1 961) and others.13 During the 1980s, hedonic regressions were

employed, for the first time, to adjust official price indices for computer equipment and new one-family

houses in the national income and product accounts of the United states. l4 Despite frequent suggestions

to the contrary, hedonic regressions do not provide any new insight into the issue of quality adjustment. The

hedonic technique is nevertheless interesting because its elaborate nature heightens difficulties which are

also implicit in other adjustment procedures.

The Promise

Like other existing approaches for quality adjustment, the hedonic regression is firmly rooted in the

assumption that quality is correlated with price. Indeed, Griliches (1 961, p 57) is resolute when he asserts

that

The reason why [at any one time] different varieties or models sell at different prices must be due to some differences in their properties, dimensions, or other "qualities,' real or imaginary. (emphasis added)

Under the common system of specification pricing, the primary focus is on a commodity and adjustment

for differences in quality are performed only irregularly. The hedonic approach, in contrast, focuses directly

on characteristics to enable a more or less continuous adjustment for quality differences. In this approach,

the familiar functional relationship between quantity (Q) and price (P) for a commodity is recast in terms of

quantity (q,) and 'implicit' prices (p,) for its n different quality dimensions, or characteristics. The general

expression for such relation can be summarized by equations (1) and (2):

(1) P = fl(Q,t),

where 'overall quantity' (Q) can be written as some aggregation of n different qualities:

(2) Q = f*(ql,...,qn).

In equation (I), the parameter associated with t is the 'pure' price change that occurs over time, while the

parameter associated with Q reflects the implicit impact of overall quantity on the commodity price. The

parameter p, associated with the characteristic q, in equation (2) Is interpreted both as the 'implicit' price

per unit of this characteristic, as well as the weight or 'contribution' of that characteristic to the overall quality

of the commodity. This simple framework seems to provide a systematic alternative to other, apparently

more erratic, quality adjustments discussed in the previous section. A closer examination reveals, however,

that this framework does not resolve any of the difficulties haunting the other methods. First, we still do

not know which quality variables are 'relevant' for each situation and we have no knowledge of an

approximate relationship between such characteristics and commodity prices. Second, the emphasis on the

seemingly 'technical' nature of hedonic regressions is highly misleading and, in fact, the neoclassical

paradigm is as dominant here as elsewhere in the quality-change literature. An illustration of these issues

is provided by the work undertaken by Griliches (1961) on automobiles.

A Studv of Automobiles

Griliches examined data for U.S. passenger fourdoor sedans for the years 1937, 1950 and 1954

through 1960. He related list prices for different models to values for three numerical quality variables

(horsepower, weight and length), as well as to six dummy variables which indicated the presence of other

quality attributes as 'standard' features 01-8 engine, hardtop, automatic transmission, power steering, power

brakes and whether the model was considered a 'compact' car). Econometric estimations were based on

the following linear semi-logarithmic form:

(3) log p = Po + XI PI^, + u,

where P is the list price for the model, q, is the value for the ith quality characteristic, p, is the implicit price

for that characteristic, po is the 'pure' price (equal for all different models) and u is an error term. This

functional form was applied to (1) cross-sectional data for each individual year, and (2) cross-sectional data

with observations drawn from two or more adjacent years. (In the second formulation, shift variables were

added to the regression in order to capture the pure price change that occurred between adjacent years.)

Hence, the first scheme allows implicit price estimates to vary from year to year, while the second assumes

that these implicit prices are fixed for the entire period.

Clearly, parameter estimates derived in this context depend crucially on the particular choice of

included variables, functional form and cross-sectional method of estimation being employed. Griliches (p.

53) admits that 'Flhere is no a priori reason to expect price and quality to be related in any particular fixed

fashion,' so he cannot explain why any specific scheme is to be preferred over alternative ones. According

to Griliches, his own decisions were based on 'empirical' considerations. A large number of different

regressions were computed for different combinations of years and independent variables, and the final

choice was based on 'an inspection of the data and the convenience of this particular formulation.' In a later

article (Griliches, 1971, p. 15) it was suggested that one should chose the framework that provided the 'most

concise and stable explanation of reality.' But reliance on 'empiricism' here could be quite precarious for

two reasons. First, there are no objective criteria for choosing the most 'convenient,' 'concise,' or 'stable'

explanation of reality among competing econometric formulations. Second, the measurement of quality

cannot depend on the explanation of price changes and remain theory-neutral at the same time. It is fairly

clear that despite its apparent rigour, the decomposition of price into 'quality' and 'pure price' dements in

the hedonic regression is far from being 'objective.' Like simpler methods, the hedonic regression also

involves a mixture of theoretical and arbitrary decisions.

To study these difficulties further, let us consider the estimates for implicit prices obtain by Griliches

and reported in his Table 3.4 (p. 66). The most striking feature here is the marked variability of coefficient

estimates between the different periods. For instance, in 1954-5, the 'implicit price' of 10 horsepower units

amounted to 2.4 per cent of the overall automobile value. In 1957-8, however, this fell to 0.4 per cent, only

to rise again in 1959-60 to 1.1 per cent. A similar variability is evident for most other parameter estimates.

Griliches (p. 64) suggests that this instability in implicit price estimates for various 'quality' attributes is an

'empirical' problem, which stems from a high correlation between the different quality variables and, also,

from a lack of variability in some quality attributes between various models. Of course, this response is not

the only possible explanation since in the hedonic framework, implicit prices should be also sensitive to

changes in the 'supply' and 'demand' functions for characteristics. Griliches (p. 79) is aware of this

possibility, which he equates with the 'classical index number problem of changing weights,' but declares

that '[Nlot much can be done about this in practice'!! For our purpose however, the issue here is not so

much the solution but the proper identification of the problem.

In the hedonic method, 'quality' is perceived to be embedded in an array of characteristics and the

hedonic regression purports to provide us with an estimated weight, or contribution of each of these

characteristics to 'overall quality.' The trouble is that these weights are not fixed, in other words, the 'quality

contents' of each characteristic changes over time. But then if the contribution of 10 horsepowers, 1 pound

of weight, or 10 inches of length to 'automobile quality' can change from year to year, this means that

quality is not solely intrinsic to the commodity and also depends on other, 'external,' factors. What are these

other factors which make quality unstable? Griliches points to changes in 'tastes' and 'supply factors' but,

in doing so, he confirms the view that quality is essentially a subjective matter. The issue is crystallised when

we note how his weights for some characteristics fluctuate between positive and negative values (p. 66). The

weight for a 100 inches of automobile length, for instance, was 0.07 per cent in 1957-8, but, in 19584, fell

to -0.18 per cent. Is it possible for an 'objective' characteristic to represent a positive quality in one year and

an 'antiquality' in the next? In light of Lancaster's treatise (1971) on objective characteristics and subjective

preferences, we suspect the answer should be negative.'' Fisher and Shell (1968, p. 24) are aiso troubled

by this conceptual riddle when they raise the basic question of 'just what we mean by taste change as

opposed to a quality change.' They provide a separate theoretical treatment for each of these concepts but,

unfortunately, they too fail to indicate how should we distinguish between quality and taste in practice.

Estimates from hedonic regressions are open to reinterpretation for other reasons. Griliches (1 961,

p. 76) writes that

One of the problems associated with the use of list prices in this study is the extent to which they may just represent pricing mistakes by manufacturers at some point in time. A manufacturer may overprice or underprice a particular innovation, and there is nothing in our method that would catch it. (emphases added)

Again, before we turn to Griliches' proposed solutions, let us contemplate the problem further. What should

we understand from Griliches' reference to 'pricing mistakes,' 'overpricing' and 'underpricing?' Clearly, these

concepts are meaningful only when we have a yardstick for 'correct' pricing, but it is common knowledge

that such an unequivocal benchmark is provided only by equilibrium in a perfectly competitive market. The

presence of this prerequisite leads to three dilemmas. (1) If pricing 'mistakes' are the result of disequilibrium,

we can no longer accept the basic assumption made in Griliches (1971, p. 4) on the existence of a

'reasonably well-fitting' relatlon between the prices of different models and the level of their characteristics.

(2) In a perfectly competitive market, prices are determined by the 'invisible hand,' not by individual

producers. Under these circumstances, it is not clear how pricing 'mistakes' by manufacturers are possible.

(3) If markets are oligopolistic or monopolistic rather than perfectly competitive, the meaning of a 'correct'

price is unclear. Should we consider a price to be the 'correct' one when it reflects the 'true quality' of the

product, when it gives rise to the maximum profit, or when company managers view it as the 'proper' one

under the circumstances? How could we know whether or not the price satisfies any one of these

requirements? Each one of these three quandaries casts heavy doubt on the validity of hedonic regressions.

Griliches prefers to ignore these questions because they are 'general:'

mhe doubt whether the evidence of the marketplace reflects adequately, if at all, the "true" marginal utility of different items or qualities to the consumer can be turned against any other price or commodity. It is not a problem peculiar to the measurement of "quality." (pp. 60-1 )

Once this is said, Griliches feels free to proceeds with suggested solutions. His remedies are interesting

because they, again, attest to strong neoclassical theoretical biases.

One sdution is to relate the temporal behaviour of residuals from the hedonic regression to the

market share for each individual model. The residuals are taken as proxies for 'pricing mistakes' as

compared to the 'true' implicit prices predicted by the regression. With this interpretation, 'overpricing' (or

'underpricing') should lead to a reduction (or an increase) in the market share of the particular model, as

consumers rearrange their purchases to maximize the quality return for their dollar. The problem with this

cure is that the proof is already embedded in the hypothesis: if producers are oligopolies, how could the

hedonic regression estimates be taken to represent the 'true' implicit price for characteristics? Furthermore,

market shares in oligopolistic markets can be influenced by a host of factors in addition to the price of

marginal utility. In particular, with massive advertising campaigns by manufacturers, one should be careful

before attributing changes in consumption patterns to 'rational' consumer decisions.

Another way of verifying the validity of coefficient estimates derived from data on new automobiles

is to compare them with quality estimates based on markets for used cars because, according to Griliches

(p. 73, 'prices of used cars are not tied any more to the manufacturers' list prices and are set, presumably,

more directly by the "market".' This alternative is deficient in two main ways. First, one may suspect that the

presence of a large oligopolistic market for new cars affects prices set in the used-car market. Second, even

if we believe that prices in this market are equilibrium outcomes of the interplay of supply and demand in

a perfectly competitive environment, we must still be able to distinguish between the price associated with

'quality' embedded in new automobiles and the discount allowed for 'depreciation' of that quality in second

hand cars. Cagan (1965) attempted to resolve this difficulty by first estimating a 'depreciation coefficient'

from market data and then adjusting quality estimates accordingly. Unfortunately, the depletion of quality

over time is no more observable than quality itself so Cagan's estimates of depreciation are only as good

as the arbitrary assumptions on which they are based.

These are only two ways by which one can amend reality to fit the preconceived neoclassical

framework. With sufficient resdve, many more can be devised. An alternative path is to recognize the reality

of an digopdistic market structure and evaluate the consequences for the hedonic-regression method.

Dhrymes (1 971), for instance, estimated implicit prices for characteristics of automobiles and refrigerators

and found that these coefficients varied between the different manufacturers. This led him (p. 104) to

conclude that the estimated coefficients represented

the manufacturer's own evaluation of the model's features in the context of his own price- quantity selling strategy - remembering that we deal essentially with an oligopolistic market and that the oligopdist may well be 'satisficing' his profits on a cost-plus-markup basis.

Under these conditions, writes Dhrymes (p. 93), 'we cannot, strictly speaking, construct 'quality corrected"

price indices routinely in the manner suggested by Court (1939) and Griliches (1961).'

We have argued that the measurement of price and quantity indices is neither objective nor free of

theoretical biases. The attempt to develop 'objective' commodity measures in the presence of quality change

is besieged by a constant resort - explicit or implicit - to 'subjective' considerations. Both the idea that

quality can be measured (objectively or not) and the methods developed for that purpose are closely tied

with the neoclassical paradigm. The evidence supporting these conclusions seems overwhelming.

This view may prove disheartening for those who prefer to have 'reliable' and 'objective' data, with

which they can describe the world and rigorously test their theories. One may hope that, although there are

serious methodological difficulties, they present only minor practical problems. Triplett (1 975), for instance,

surveyed numerous studies on quality changes and concluded that there are no clear evidence that price

indices are systematically biased in one way or the other. This conclusion, in itself, reflects a

misunderstanding of the problem. Such inference is possible only if we already assumed that price indices

can, in fact, be denominated in some neutral, a-historic units of 'quality.' The nature of this problem is best

illustrated in reference to long-term historical comparisons. Economists often examine price and quantity

series that extend over a century or more but the meaning of such comparisons is unclear. For instance,

how should we interpret the measure of real GNP in 1882 when denominated in '1982 prices?' Most

commodities produced in 1882 were simply unavailable in 1982 and hence could not have '1982 prices.'

Furthermore, every generation of goods and services introduces new features and eliminates older ones,

so even the imputation of implicit prices for characteristics is infeasible. Finally, the 'desirability' of any

commodity attribute and the 'satisfaction' it may provide changes drastically with social conditions. In this

context of vast changes, comparisons of uniform quality units may often seem absurd but the United

Nations' System of National Accounts (1968, p. 61) nevertheless hopes to circumscribe the problem by

arresting history in a straightjacket. It suggests to formulate characteristics so that 'their validity is as far as

possible timeless.'

Is it at all possible to have theory-neutral, objective indices for price and quantity in the presence

of historical change? In our view, the answer is negative but this need not destroy the prospect for empirical

economic research. Like Carr (1 961 ), we may also believe that any study of society necessarily reflects

subjective biases of the observer and that such biases must be recognized, not ignored.

Notes

1. In the first case, when 'automobile quality' is doubled, a purchase of one 1985 Mustang for $14,000 is equivalent to buying two 1975 models for a unit price of $7,000. This imputed price is 30 per cent lower than the 1975 price of $10,000. The doubling of quality also implies that output (denominated in 1975 units) rose from 100,000 in 1975 to 300,000 (1 50,000 x 2) in 1985, or a 200 per cent increase. In the second case, when 'automobile quality' is halved, the imputed price for a 1985 equivalent of one 1975 Mustang rises to $28,000 ($14,000 x 2), an increase of 180 per cent over the original 1975 unit price. Quantity (denominated in 1975 units) falls to 75,000 (150,000 + 2), or a decrease of 25 per cent from the 1975 level.

2. Ever since the Greek epistemology separated the earthly appearance of things from the universal truth of the world spirit, philosophers have attempted to bridge this gap between the subjective and the objective. British empiricists like Locke and Hume rejected the notion of universality and argued that apparent uniformity in human thinking was rooted only in customs and habits. German idealists, like Kant, contested this view and argued that perception was universal - namely common to all humans - because it required the a priori intuition of space and time and various categories such as unity, substantiality and causality. Nevertheless, the Platonic spirit continued to fly unarrested for Kant agreed that objects of experience were indeed never given by themselves and could only be perceived through human experience. Even Hegel's later attempt to unify object and subject into one 'whole' failed to resolve the riddle.

3. This view is widely held. For similar expressions, see Adelman and Griliches (1961, p. 539), Court (1 939, p. 107-8), Rosen (1974, pp. 75-6) and Triplett (1983, pp. 277-8; 1986, p. 37).

4. This distinction was also central to the earlier work by Fisher and Shell (1972).

5. For recognition of these prerequisites, see Early and Sinclair (1983, p. 108), Hofsten (1 949, pp. 285- 6), Rosen (1974), Ulmer (1949, pp. 67-8).

6. There is another reason for concern here. For consumer goods, for instance, the choice of relevant characteristics in this method depends on utility functions and consumer preferences. But how can one rely on subjective consumer preferences to describe 'objective' characteristics of commodities? We return to this important question is the fourth section.

7. For example, see Triplett (1983, p. 274 and p. 302).

8. On the lack of such narratives, see comments made in the Price Statistics Review Committee (1961) and by Griliches (1 967 and 1971).

9. For further details on quality adjustment to the PPI, see Early and Sinclair (1983).

10. For an explanation of 'spiicing', see The United Nations (1977), p. 9.

11. The process is not as simple if the older turntable disappears from the market before the new CD player is introduced into the CPI. When this happens, the price of for the non-existing turntable that 'would have prevailed' in the time of the swap must be somehow imputed. A similar difficulty arises when a totally new product is introduced into the index.

12. For further details on the construction of price indices for military spending, see Ziemer and Galbraith (1 983).

13. For a detail bibliography on 'hedonic regressions', see Griliches (ed.) (1971) and Triplett (1975). References to recent studies are cited in Cole et a/. (1986) and Triplett (1986).

14. See Cole et a/. (1986) and Triplett (1 986) for more information.

15. As we argued, Lancaster's taxonomy contains some serious inconsistencies and ambiguities so the answer here cannot be definite.

Adelman, I. and Z. Griliches (1 961 ), 'On the Index of Quality Change,' Journal of the American Statistical Association, September, pp. 535-48.

Cagan, P. (1965, 1971), 'Measuring Quality Changes and the Purchasing Power of Money: an Explanatory Study of Automobiles,' ch. 7 in Z. Griliches (4.).

Carr, E.H. (1961), What is Histog0 (Harmondsworth, Middlesex: Penguin). The George Macaulay Trevelyn Lectures, ddivered at the University of Cambridge.

Cole, R., Y.C. Chen, J.A. Barquin-Stolleman, E. Dulberger, N. Hdvacian and J.H. Hodge (1986), 'Quality Adjusted Price lndexes for Computer Processors and Sdected Peripheral Equipment,' in U.S. Bureau of Economic Analysis, Survey of Current Business, January, pp. 41 -50.

Court, A.T. (1939), 'Hedonic Price lndexes with Automotive Examples,' in The Dynamics of Automobile Demand (New York: General Motors Corporation).

Dhrymes, P.J. (1971), 'Price and Quality Changes in Consumer Capital Goods: An Empirical Study,' ch. 4 in Z. Griliches (4.).

Early, J.F. and J.H. Sinclair (1983), 'Quality Adjustment in the Producer Price Indexes,' ch. 2 in M.F. Foss (4.).

Fisher, F.M. and K. Shell (1968, 1971), 'Tastes and Quality Change in the Pure Theory of the True-Cost-of- Living Index,' ch 2 in Z. Griliches (4.).

--. (1 972), The Economic Theory of Price Indices (New York: Academic Press).

Foss, M.F. (4.) (1 983), The U.S. National Income and Product Accounts: Selected Topics, National Bureau of Economic Research, Studies in Income and Wealth, Vd. 47, (Chicago: The University of Chicago Press) .

Griliches, Z. (1 961, 1971), 'Hedonic Price lndexes for Autornobiles: An Econometric Analysis of Quality Change,' ch. 3 in Z. Griliches (4.).

---. (1 971), 'Introduction: Hedonic Price lndexes Revisited,' ch. 1 in Z. Griliches (4.).

Griliches, Z. (4.) (1 971), Price lndexes and Quality Change. Studies in New Methods of Measurement (Carnbridge, Mass.: Harvard University Press).

Hofsten, E.V. (1949), 'Construction of Consumers' Price Indexes, Technical Notes,' Monthly Labor Review, Vol. 69, No. 3, September.

Huxley, A. (1 932), Brave New World (New York: The Modern Library).

Lancaster, K. (1 971), Consumer Demand. A New Approach (New York and London: Columbia University Press) .

Price Statistics Review Committee (1961), The Price Statistics of the Federal Government. Review Appraisal and Recommendations (New York: National Bureau of Economic Research).

Rosen, S. (1 974), 'Hedonic Prices and Implicit Markets: Product Differentiation in Pure competition,' Journal of Political Economy, No. 82, January-February, pp. 34-55.

Stone, R. (1956), Quality and Price lndexes in National Accounts (Paris: Organization for European Economic Co-operation).

Triplet, J.E. (1971), 'Quality Bias in Price lndexes and New Methods of Quality Measurement,' ch. 6 in Z. Griliches (ed.).

--. (1 975), 'The Measurement of Inflation: A Survey of Research on the Accuracy of Price Indexes, in P.H. Earl (ed.) Analysis of Inflation (Lexington, Mass.: Lexington Books, D.C. Heath and Company).

---. (I-), 'Concepts of Quality in Input and Output Price Measures: A Resolution of the User-Value Resource-Cost Debate,' ch.5 in M.F. Foss (ed.).

--. (1 986), 'The Economic Interpretation of Hedonic Methods,' in U.S. Bureau of Economic Analysis, Survey of Current Business, Vd. 66, No. 1, January, pp. 36-40.

Ulmer, M. J. (1 949), The Economic Theory of Cost of Living Index Numbers (New York: Columbia University Press).

United Nations, Department of Economic and Social Affairs, Statistical Office (1968), A System of National Accounts, Studies in Methods, Series F, No. 2, Rev. 3 (New York: United Nations).

--. (1 977), Guidelines on Principles of a System of Price and Quantity Statistics, Statistical Papers, Series M, No. 59 (New York: United Nations).

Ziemer, R. and K.D. Galbraith (I-), 'Deflation of Defense Purchases,' in M.F. Foss (ed.).